Ψ-0.6 unified vision-language world model reasons in multimodal streams

Researchers released Ψ-0.6, a single model that reasons through streams spanning words, depth, motion, camera pose and imagined scenes, with RL learning which representation to use. It reportedly beats leading open-weight models on zero-shot MindCube spatial reasoning and outperforms other world models on FlowBench, transferring from simulated rooms to real robot tabletops.